The misleading timeline is the issue, not the new idea
Norbert Kerr defined HARKing as presenting a post-hoc hypothesis informed by the results as if it were an a-priori hypothesis. Generating an explanation after inspecting observations is ordinary exploratory work. The problem is claiming that this explanation guided the design or analysis before the pattern was available. That account hides how the hypothesis and evidence became dependent.
A dissertation can investigate unexpected findings. It should preserve the hypothesis stated in the proposal or analysis plan, record when the outcomes were viewed and label the subsequent idea. Clear chronology helps readers judge evidential status and prevents the author’s own memory from turning discovery into prediction.
Exploration and confirmation carry different uncertainty
A planned test normally fixes the hypothesis, variables, direction, population, exclusions and analysis before the target result is inspected. Exploratory work may examine many patterns. The selected finding can produce a strong hypothesis, but it also carries the flexibility of the search. The uncertainty calculation for one final model does not necessarily represent all the unreported routes used to select it.
If the selected route is described as a targeted prediction, readers cannot see how many alternatives were considered or discarded. Confidence may then be calibrated to a cleaner process than the one conducted. The answer is not to suppress exploration. It is to report the search, preserve the original tests and, where possible, evaluate the new hypothesis on independent observations.
A change of plan is not automatically HARKing
Research needs adjustment. An instrument may fail, model assumptions may be untenable, missingness may differ from expectations, or an input error may be discovered. Preserve the initial plan, state the trigger, record whether results were known and explain the replacement decision. A justified departure remains credible when its origin and impact are visible.
Distinguish outcome-independent and outcome-dependent reasons. Discovering a coding error before the target analysis differs from switching models after an unwelcome result. An outcome-dependent follow-up can still be informative, but it must be classified as exploratory or as a sensitivity analysis with its trigger stated. It cannot be narrated as an untouched original test.
Use a hypothesis register instead of memory
Maintain a table containing hypothesis ID, exact wording, theoretical basis, outcome, direction, model, creation date and data status. Never overwrite an old row; issue a new version and state which data had already been seen. Link each hypothesis to its code and output. Include null or unsupported planned hypotheses rather than allowing them to disappear from the final story.
Formal preregistration can provide a time-stamped record but is not the only possible evidence. A versioned proposal, signed analysis sheet or dated supervisory handover may suit student work where institutional rules allow it. The content must be specific. A timestamp attached to a vague intention does not establish a detailed confirmatory plan.
Turn discovery into an honest follow-up question
Report the planned route and unsupported result first. Follow it with a clearly labelled exploratory section. State which variables, subgroups and models were inspected and why the selected pattern deserves attention. Phrase the new hypothesis as something to test in future, not an explanation already proved by the observations from which it was constructed.
A useful form is: “The pre-specified association A was not supported. Exploratory analysis indicated pattern B. Because B was selected after inspection, we treat it as hypothesis-generating. An independent study should specify the variables, direction and model in advance.” Describe magnitude and uncertainty as well as any significance label.
Repair a manuscript whose chronology has been polished away
Compare the introduction and method with early notes, proposals, file versions and messages. Highlight hypotheses without a contemporaneous record. Do not simply move them into an earlier section. Restore the original plan, label later accounts and document departures. Where timing cannot be reconstructed, exploratory is the more defensible status.
If the correction changes a central conclusion, discuss it with the supervisor and follow the institution’s procedure. Correcting the report is not itself an admission of formal misconduct. This page cannot determine such a finding; context, intent and the authorised institutional process matter.
Pre-submission HARKing audit
- Does every hypothesis described as prior have a dated, substantive record?
- Do unsupported original hypotheses remain visible?
- Are departures documented with timing, information state and rationale?
- Is the scope and uncertainty of exploratory work reported?
- Are newly generated hypotheses framed for independent future testing?